The SageMaker Object Detection algorithm identifies and classifies objects in images. The identified object is placed in a class with a numerical measure of confidence. The location in the image is identified by a bounding box around the object. Object Detection is a Supervised Learning algorithm trained on a corpus of labeled images.
Because the Object Detection algorithm returns the location of the object on the image it is possible to process the output further. This leads to richer use cases where more information is extracted from the object. Examples include reading bar codes, identifying product items and determining the state of an object i.e. is it defective, or unsafe.
- AWS docs: https://docs.aws.amazon.com/sagemaker/latest/dg/object-detection.html
- AWS blog: https://aws.amazon.com/blogs/iot/sagemaker-object-detection-greengrass-part-1-of-3/
|Data types and format||Image|
|Learning paradigm or domain||Image Processing, Supervised Learning|
|Problem type||Object detection and classification|
|Use case examples||Detect people and objects in an image|
The algorithm can be trained from scratch or by models from pre-trained on the ImageNet data. The recommended format for training data is the Apache MxNet recordIO format, although it will also accept jpeg and png.
To speed up training you can seed the training data with data from a model you trained previously. This is called incremental training.
Model artifacts and inference
|Learning paradigm||Supervised Learning|
|Request format||Recommended: application/x-image; Also jpeg, png|
Both CPU and GPU instances can be used in single or multi-instance configurations. The GPU instance can have multiple GPUs.
AWS AI-ML Partner Deep Dive Webinar: Object Detection and Image Classification Algorithms
This is a one hour and 9 minute video from AWS.